src/gromacs/gmxlib/nonbonded/nb_kernel_sse4_1_single/nb_kernel_ElecEw_VdwCSTab_GeomW3P1_sse4_1_single.c

   1 /*
   2  * This file is part of the GROMACS molecular simulation package.
   3  *
   4  * Copyright (c) 2012,2013, by the GROMACS development team, led by
   5  * Mark Abraham, David van der Spoel, Berk Hess, and Erik Lindahl,
   6  * and including many others, as listed in the AUTHORS file in the
   7  * top-level source directory and at http://www.gromacs.org.
   8  *
   9  * GROMACS is free software; you can redistribute it and/or
  10  * modify it under the terms of the GNU Lesser General Public License
  11  * as published by the Free Software Foundation; either version 2.1
  12  * of the License, or (at your option) any later version.
  13  *
  14  * GROMACS is distributed in the hope that it will be useful,
  15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
  16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
  17  * Lesser General Public License for more details.
  18  *
  19  * You should have received a copy of the GNU Lesser General Public
  20  * License along with GROMACS; if not, see
  21  * http://www.gnu.org/licenses, or write to the Free Software Foundation,
  22  * Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA.
  23  *
  24  * If you want to redistribute modifications to GROMACS, please
  25  * consider that scientific software is very special. Version
  26  * control is crucial - bugs must be traceable. We will be happy to
  27  * consider code for inclusion in the official distribution, but
  28  * derived work must not be called official GROMACS. Details are found
  29  * in the README & COPYING files - if they are missing, get the
  30  * official version at http://www.gromacs.org.
  31  *
  32  * To help us fund GROMACS development, we humbly ask that you cite
  33  * the research papers on the package. Check out http://www.gromacs.org.
  34  */
  35 /*
  36  * Note: this file was generated by the GROMACS sse4_1_single kernel generator.
  37  */
  38 #ifdef HAVE_CONFIG_H
  39 #include <config.h>
  40 #endif
  41
  42 #include <math.h>
  43
  44 #include "../nb_kernel.h"
  45 #include "types/simple.h"
  46 #include "vec.h"
  47 #include "nrnb.h"
  48
  49 #include "gmx_math_x86_sse4_1_single.h"
  50 #include "kernelutil_x86_sse4_1_single.h"
  51
  52 /*
  53  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_single
  54  * Electrostatics interaction: Ewald
  55  * VdW interaction:            CubicSplineTable
  56  * Geometry:                   Water3-Particle
  57  * Calculate force/pot:        PotentialAndForce
  58  */
  59 void
  60 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_VF_sse4_1_single
  61                     (t_nblist                    * gmx_restrict       nlist,
  62                      rvec                        * gmx_restrict          xx,
  63                      rvec                        * gmx_restrict          ff,
  64                      t_forcerec                  * gmx_restrict          fr,
  65                      t_mdatoms                   * gmx_restrict     mdatoms,
  66                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
  67                      t_nrnb                      * gmx_restrict        nrnb)
  68 {
  69     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
  70      * just 0 for non-waters.
  71      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
  72      * jnr indices corresponding to data put in the four positions in the SIMD register.
  73      */
  74     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
  75     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
  76     int              jnrA,jnrB,jnrC,jnrD;
  77     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
  78     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
  79     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
  80     real             rcutoff_scalar;
  81     real             *shiftvec,*fshift,*x,*f;
  82     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
  83     real             scratch[4*DIM];
  84     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
  85     int              vdwioffset0;
  86     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
  87     int              vdwioffset1;
  88     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
  89     int              vdwioffset2;
  90     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
  91     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
  92     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
  93     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
  94     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
  95     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
  96     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
  97     real             *charge;
  98     int              nvdwtype;
  99     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 100     int              *vdwtype;
 101     real             *vdwparam;
 102     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 103     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 104     __m128i          vfitab;
 105     __m128i          ifour       = _mm_set1_epi32(4);
 106     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 107     real             *vftab;
 108     __m128i          ewitab;
 109     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 110     real             *ewtab;
 111     __m128           dummy_mask,cutoff_mask;
 112     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 113     __m128           one     = _mm_set1_ps(1.0);
 114     __m128           two     = _mm_set1_ps(2.0);
 115     x                = xx[0];
 116     f                = ff[0];
 117
 118     nri              = nlist->nri;
 119     iinr             = nlist->iinr;
 120     jindex           = nlist->jindex;
 121     jjnr             = nlist->jjnr;
 122     shiftidx         = nlist->shift;
 123     gid              = nlist->gid;
 124     shiftvec         = fr->shift_vec[0];
 125     fshift           = fr->fshift[0];
 126     facel            = _mm_set1_ps(fr->epsfac);
 127     charge           = mdatoms->chargeA;
 128     nvdwtype         = fr->ntype;
 129     vdwparam         = fr->nbfp;
 130     vdwtype          = mdatoms->typeA;
 131
 132     vftab            = kernel_data->table_vdw->data;
 133     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 134
 135     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 136     ewtab            = fr->ic->tabq_coul_FDV0;
 137     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 138     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 139
 140     /* Setup water-specific parameters */
 141     inr              = nlist->iinr[0];
 142     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 143     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 144     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 145     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 146
 147     /* Avoid stupid compiler warnings */
 148     jnrA = jnrB = jnrC = jnrD = 0;
 149     j_coord_offsetA = 0;
 150     j_coord_offsetB = 0;
 151     j_coord_offsetC = 0;
 152     j_coord_offsetD = 0;
 153
 154     outeriter        = 0;
 155     inneriter        = 0;
 156
 157     for(iidx=0;iidx<4*DIM;iidx++)
 158     {
 159         scratch[iidx] = 0.0;
 160     }
 161
 162     /* Start outer loop over neighborlists */
 163     for(iidx=0; iidx<nri; iidx++)
 164     {
 165         /* Load shift vector for this list */
 166         i_shift_offset   = DIM*shiftidx[iidx];
 167
 168         /* Load limits for loop over neighbors */
 169         j_index_start    = jindex[iidx];
 170         j_index_end      = jindex[iidx+1];
 171
 172         /* Get outer coordinate index */
 173         inr              = iinr[iidx];
 174         i_coord_offset   = DIM*inr;
 175
 176         /* Load i particle coords and add shift vector */
 177         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 178                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 179
 180         fix0             = _mm_setzero_ps();
 181         fiy0             = _mm_setzero_ps();
 182         fiz0             = _mm_setzero_ps();
 183         fix1             = _mm_setzero_ps();
 184         fiy1             = _mm_setzero_ps();
 185         fiz1             = _mm_setzero_ps();
 186         fix2             = _mm_setzero_ps();
 187         fiy2             = _mm_setzero_ps();
 188         fiz2             = _mm_setzero_ps();
 189
 190         /* Reset potential sums */
 191         velecsum         = _mm_setzero_ps();
 192         vvdwsum          = _mm_setzero_ps();
 193
 194         /* Start inner kernel loop */
 195         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 196         {
 197
 198             /* Get j neighbor index, and coordinate index */
 199             jnrA             = jjnr[jidx];
 200             jnrB             = jjnr[jidx+1];
 201             jnrC             = jjnr[jidx+2];
 202             jnrD             = jjnr[jidx+3];
 203             j_coord_offsetA  = DIM*jnrA;
 204             j_coord_offsetB  = DIM*jnrB;
 205             j_coord_offsetC  = DIM*jnrC;
 206             j_coord_offsetD  = DIM*jnrD;
 207
 208             /* load j atom coordinates */
 209             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 210                                               x+j_coord_offsetC,x+j_coord_offsetD,
 211                                               &jx0,&jy0,&jz0);
 212
 213             /* Calculate displacement vector */
 214             dx00             = _mm_sub_ps(ix0,jx0);
 215             dy00             = _mm_sub_ps(iy0,jy0);
 216             dz00             = _mm_sub_ps(iz0,jz0);
 217             dx10             = _mm_sub_ps(ix1,jx0);
 218             dy10             = _mm_sub_ps(iy1,jy0);
 219             dz10             = _mm_sub_ps(iz1,jz0);
 220             dx20             = _mm_sub_ps(ix2,jx0);
 221             dy20             = _mm_sub_ps(iy2,jy0);
 222             dz20             = _mm_sub_ps(iz2,jz0);
 223
 224             /* Calculate squared distance and things based on it */
 225             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 226             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 227             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 228
 229             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 230             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 231             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 232
 233             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 234             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 235             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 236
 237             /* Load parameters for j particles */
 238             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 239                                                               charge+jnrC+0,charge+jnrD+0);
 240             vdwjidx0A        = 2*vdwtype[jnrA+0];
 241             vdwjidx0B        = 2*vdwtype[jnrB+0];
 242             vdwjidx0C        = 2*vdwtype[jnrC+0];
 243             vdwjidx0D        = 2*vdwtype[jnrD+0];
 244
 245             fjx0             = _mm_setzero_ps();
 246             fjy0             = _mm_setzero_ps();
 247             fjz0             = _mm_setzero_ps();
 248
 249             /**************************
 250              * CALCULATE INTERACTIONS *
 251              **************************/
 252
 253             r00              = _mm_mul_ps(rsq00,rinv00);
 254
 255             /* Compute parameters for interactions between i and j atoms */
 256             qq00             = _mm_mul_ps(iq0,jq0);
 257             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 258                                          vdwparam+vdwioffset0+vdwjidx0B,
 259                                          vdwparam+vdwioffset0+vdwjidx0C,
 260                                          vdwparam+vdwioffset0+vdwjidx0D,
 261                                          &c6_00,&c12_00);
 262
 263             /* Calculate table index by multiplying r with table scale and truncate to integer */
 264             rt               = _mm_mul_ps(r00,vftabscale);
 265             vfitab           = _mm_cvttps_epi32(rt);
 266             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 267             vfitab           = _mm_slli_epi32(vfitab,3);
 268
 269             /* EWALD ELECTROSTATICS */
 270
 271             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 272             ewrt             = _mm_mul_ps(r00,ewtabscale);
 273             ewitab           = _mm_cvttps_epi32(ewrt);
 274             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 275             ewitab           = _mm_slli_epi32(ewitab,2);
 276             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 277             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 278             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 279             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 280             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 281             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 282             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 283             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 284             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 285
 286             /* CUBIC SPLINE TABLE DISPERSION */
 287             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 288             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 289             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 290             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 291             _MM_TRANSPOSE4_PS(Y,F,G,H);
 292             Heps             = _mm_mul_ps(vfeps,H);
 293             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 294             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 295             vvdw6            = _mm_mul_ps(c6_00,VV);
 296             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 297             fvdw6            = _mm_mul_ps(c6_00,FF);
 298
 299             /* CUBIC SPLINE TABLE REPULSION */
 300             vfitab           = _mm_add_epi32(vfitab,ifour);
 301             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 302             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 303             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 304             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 305             _MM_TRANSPOSE4_PS(Y,F,G,H);
 306             Heps             = _mm_mul_ps(vfeps,H);
 307             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 308             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 309             vvdw12           = _mm_mul_ps(c12_00,VV);
 310             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 311             fvdw12           = _mm_mul_ps(c12_00,FF);
 312             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 313             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 314
 315             /* Update potential sum for this i atom from the interaction with this j atom. */
 316             velecsum         = _mm_add_ps(velecsum,velec);
 317             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 318
 319             fscal            = _mm_add_ps(felec,fvdw);
 320
 321             /* Calculate temporary vectorial force */
 322             tx               = _mm_mul_ps(fscal,dx00);
 323             ty               = _mm_mul_ps(fscal,dy00);
 324             tz               = _mm_mul_ps(fscal,dz00);
 325
 326             /* Update vectorial force */
 327             fix0             = _mm_add_ps(fix0,tx);
 328             fiy0             = _mm_add_ps(fiy0,ty);
 329             fiz0             = _mm_add_ps(fiz0,tz);
 330
 331             fjx0             = _mm_add_ps(fjx0,tx);
 332             fjy0             = _mm_add_ps(fjy0,ty);
 333             fjz0             = _mm_add_ps(fjz0,tz);
 334
 335             /**************************
 336              * CALCULATE INTERACTIONS *
 337              **************************/
 338
 339             r10              = _mm_mul_ps(rsq10,rinv10);
 340
 341             /* Compute parameters for interactions between i and j atoms */
 342             qq10             = _mm_mul_ps(iq1,jq0);
 343
 344             /* EWALD ELECTROSTATICS */
 345
 346             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 347             ewrt             = _mm_mul_ps(r10,ewtabscale);
 348             ewitab           = _mm_cvttps_epi32(ewrt);
 349             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 350             ewitab           = _mm_slli_epi32(ewitab,2);
 351             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 352             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 353             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 354             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 355             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 356             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 357             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 358             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 359             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 360
 361             /* Update potential sum for this i atom from the interaction with this j atom. */
 362             velecsum         = _mm_add_ps(velecsum,velec);
 363
 364             fscal            = felec;
 365
 366             /* Calculate temporary vectorial force */
 367             tx               = _mm_mul_ps(fscal,dx10);
 368             ty               = _mm_mul_ps(fscal,dy10);
 369             tz               = _mm_mul_ps(fscal,dz10);
 370
 371             /* Update vectorial force */
 372             fix1             = _mm_add_ps(fix1,tx);
 373             fiy1             = _mm_add_ps(fiy1,ty);
 374             fiz1             = _mm_add_ps(fiz1,tz);
 375
 376             fjx0             = _mm_add_ps(fjx0,tx);
 377             fjy0             = _mm_add_ps(fjy0,ty);
 378             fjz0             = _mm_add_ps(fjz0,tz);
 379
 380             /**************************
 381              * CALCULATE INTERACTIONS *
 382              **************************/
 383
 384             r20              = _mm_mul_ps(rsq20,rinv20);
 385
 386             /* Compute parameters for interactions between i and j atoms */
 387             qq20             = _mm_mul_ps(iq2,jq0);
 388
 389             /* EWALD ELECTROSTATICS */
 390
 391             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 392             ewrt             = _mm_mul_ps(r20,ewtabscale);
 393             ewitab           = _mm_cvttps_epi32(ewrt);
 394             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 395             ewitab           = _mm_slli_epi32(ewitab,2);
 396             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 397             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 398             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 399             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 400             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 401             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 402             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 403             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 404             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 405
 406             /* Update potential sum for this i atom from the interaction with this j atom. */
 407             velecsum         = _mm_add_ps(velecsum,velec);
 408
 409             fscal            = felec;
 410
 411             /* Calculate temporary vectorial force */
 412             tx               = _mm_mul_ps(fscal,dx20);
 413             ty               = _mm_mul_ps(fscal,dy20);
 414             tz               = _mm_mul_ps(fscal,dz20);
 415
 416             /* Update vectorial force */
 417             fix2             = _mm_add_ps(fix2,tx);
 418             fiy2             = _mm_add_ps(fiy2,ty);
 419             fiz2             = _mm_add_ps(fiz2,tz);
 420
 421             fjx0             = _mm_add_ps(fjx0,tx);
 422             fjy0             = _mm_add_ps(fjy0,ty);
 423             fjz0             = _mm_add_ps(fjz0,tz);
 424
 425             fjptrA             = f+j_coord_offsetA;
 426             fjptrB             = f+j_coord_offsetB;
 427             fjptrC             = f+j_coord_offsetC;
 428             fjptrD             = f+j_coord_offsetD;
 429
 430             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 431
 432             /* Inner loop uses 157 flops */
 433         }
 434
 435         if(jidx<j_index_end)
 436         {
 437
 438             /* Get j neighbor index, and coordinate index */
 439             jnrlistA         = jjnr[jidx];
 440             jnrlistB         = jjnr[jidx+1];
 441             jnrlistC         = jjnr[jidx+2];
 442             jnrlistD         = jjnr[jidx+3];
 443             /* Sign of each element will be negative for non-real atoms.
 444              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
 445              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
 446              */
 447             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
 448             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
 449             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
 450             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
 451             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
 452             j_coord_offsetA  = DIM*jnrA;
 453             j_coord_offsetB  = DIM*jnrB;
 454             j_coord_offsetC  = DIM*jnrC;
 455             j_coord_offsetD  = DIM*jnrD;
 456
 457             /* load j atom coordinates */
 458             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 459                                               x+j_coord_offsetC,x+j_coord_offsetD,
 460                                               &jx0,&jy0,&jz0);
 461
 462             /* Calculate displacement vector */
 463             dx00             = _mm_sub_ps(ix0,jx0);
 464             dy00             = _mm_sub_ps(iy0,jy0);
 465             dz00             = _mm_sub_ps(iz0,jz0);
 466             dx10             = _mm_sub_ps(ix1,jx0);
 467             dy10             = _mm_sub_ps(iy1,jy0);
 468             dz10             = _mm_sub_ps(iz1,jz0);
 469             dx20             = _mm_sub_ps(ix2,jx0);
 470             dy20             = _mm_sub_ps(iy2,jy0);
 471             dz20             = _mm_sub_ps(iz2,jz0);
 472
 473             /* Calculate squared distance and things based on it */
 474             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 475             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 476             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 477
 478             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 479             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 480             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 481
 482             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 483             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 484             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 485
 486             /* Load parameters for j particles */
 487             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 488                                                               charge+jnrC+0,charge+jnrD+0);
 489             vdwjidx0A        = 2*vdwtype[jnrA+0];
 490             vdwjidx0B        = 2*vdwtype[jnrB+0];
 491             vdwjidx0C        = 2*vdwtype[jnrC+0];
 492             vdwjidx0D        = 2*vdwtype[jnrD+0];
 493
 494             fjx0             = _mm_setzero_ps();
 495             fjy0             = _mm_setzero_ps();
 496             fjz0             = _mm_setzero_ps();
 497
 498             /**************************
 499              * CALCULATE INTERACTIONS *
 500              **************************/
 501
 502             r00              = _mm_mul_ps(rsq00,rinv00);
 503             r00              = _mm_andnot_ps(dummy_mask,r00);
 504
 505             /* Compute parameters for interactions between i and j atoms */
 506             qq00             = _mm_mul_ps(iq0,jq0);
 507             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 508                                          vdwparam+vdwioffset0+vdwjidx0B,
 509                                          vdwparam+vdwioffset0+vdwjidx0C,
 510                                          vdwparam+vdwioffset0+vdwjidx0D,
 511                                          &c6_00,&c12_00);
 512
 513             /* Calculate table index by multiplying r with table scale and truncate to integer */
 514             rt               = _mm_mul_ps(r00,vftabscale);
 515             vfitab           = _mm_cvttps_epi32(rt);
 516             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 517             vfitab           = _mm_slli_epi32(vfitab,3);
 518
 519             /* EWALD ELECTROSTATICS */
 520
 521             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 522             ewrt             = _mm_mul_ps(r00,ewtabscale);
 523             ewitab           = _mm_cvttps_epi32(ewrt);
 524             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 525             ewitab           = _mm_slli_epi32(ewitab,2);
 526             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 527             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 528             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 529             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 530             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 531             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 532             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 533             velec            = _mm_mul_ps(qq00,_mm_sub_ps(rinv00,velec));
 534             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 535
 536             /* CUBIC SPLINE TABLE DISPERSION */
 537             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 538             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 539             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 540             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 541             _MM_TRANSPOSE4_PS(Y,F,G,H);
 542             Heps             = _mm_mul_ps(vfeps,H);
 543             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 544             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 545             vvdw6            = _mm_mul_ps(c6_00,VV);
 546             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 547             fvdw6            = _mm_mul_ps(c6_00,FF);
 548
 549             /* CUBIC SPLINE TABLE REPULSION */
 550             vfitab           = _mm_add_epi32(vfitab,ifour);
 551             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 552             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 553             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 554             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 555             _MM_TRANSPOSE4_PS(Y,F,G,H);
 556             Heps             = _mm_mul_ps(vfeps,H);
 557             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 558             VV               = _mm_add_ps(Y,_mm_mul_ps(vfeps,Fp));
 559             vvdw12           = _mm_mul_ps(c12_00,VV);
 560             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 561             fvdw12           = _mm_mul_ps(c12_00,FF);
 562             vvdw             = _mm_add_ps(vvdw12,vvdw6);
 563             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 564
 565             /* Update potential sum for this i atom from the interaction with this j atom. */
 566             velec            = _mm_andnot_ps(dummy_mask,velec);
 567             velecsum         = _mm_add_ps(velecsum,velec);
 568             vvdw             = _mm_andnot_ps(dummy_mask,vvdw);
 569             vvdwsum          = _mm_add_ps(vvdwsum,vvdw);
 570
 571             fscal            = _mm_add_ps(felec,fvdw);
 572
 573             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 574
 575             /* Calculate temporary vectorial force */
 576             tx               = _mm_mul_ps(fscal,dx00);
 577             ty               = _mm_mul_ps(fscal,dy00);
 578             tz               = _mm_mul_ps(fscal,dz00);
 579
 580             /* Update vectorial force */
 581             fix0             = _mm_add_ps(fix0,tx);
 582             fiy0             = _mm_add_ps(fiy0,ty);
 583             fiz0             = _mm_add_ps(fiz0,tz);
 584
 585             fjx0             = _mm_add_ps(fjx0,tx);
 586             fjy0             = _mm_add_ps(fjy0,ty);
 587             fjz0             = _mm_add_ps(fjz0,tz);
 588
 589             /**************************
 590              * CALCULATE INTERACTIONS *
 591              **************************/
 592
 593             r10              = _mm_mul_ps(rsq10,rinv10);
 594             r10              = _mm_andnot_ps(dummy_mask,r10);
 595
 596             /* Compute parameters for interactions between i and j atoms */
 597             qq10             = _mm_mul_ps(iq1,jq0);
 598
 599             /* EWALD ELECTROSTATICS */
 600
 601             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 602             ewrt             = _mm_mul_ps(r10,ewtabscale);
 603             ewitab           = _mm_cvttps_epi32(ewrt);
 604             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 605             ewitab           = _mm_slli_epi32(ewitab,2);
 606             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 607             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 608             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 609             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 610             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 611             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 612             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 613             velec            = _mm_mul_ps(qq10,_mm_sub_ps(rinv10,velec));
 614             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
 615
 616             /* Update potential sum for this i atom from the interaction with this j atom. */
 617             velec            = _mm_andnot_ps(dummy_mask,velec);
 618             velecsum         = _mm_add_ps(velecsum,velec);
 619
 620             fscal            = felec;
 621
 622             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 623
 624             /* Calculate temporary vectorial force */
 625             tx               = _mm_mul_ps(fscal,dx10);
 626             ty               = _mm_mul_ps(fscal,dy10);
 627             tz               = _mm_mul_ps(fscal,dz10);
 628
 629             /* Update vectorial force */
 630             fix1             = _mm_add_ps(fix1,tx);
 631             fiy1             = _mm_add_ps(fiy1,ty);
 632             fiz1             = _mm_add_ps(fiz1,tz);
 633
 634             fjx0             = _mm_add_ps(fjx0,tx);
 635             fjy0             = _mm_add_ps(fjy0,ty);
 636             fjz0             = _mm_add_ps(fjz0,tz);
 637
 638             /**************************
 639              * CALCULATE INTERACTIONS *
 640              **************************/
 641
 642             r20              = _mm_mul_ps(rsq20,rinv20);
 643             r20              = _mm_andnot_ps(dummy_mask,r20);
 644
 645             /* Compute parameters for interactions between i and j atoms */
 646             qq20             = _mm_mul_ps(iq2,jq0);
 647
 648             /* EWALD ELECTROSTATICS */
 649
 650             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 651             ewrt             = _mm_mul_ps(r20,ewtabscale);
 652             ewitab           = _mm_cvttps_epi32(ewrt);
 653             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 654             ewitab           = _mm_slli_epi32(ewitab,2);
 655             ewtabF           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,0) );
 656             ewtabD           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,1) );
 657             ewtabV           = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,2) );
 658             ewtabFn          = _mm_load_ps( ewtab + gmx_mm_extract_epi32(ewitab,3) );
 659             _MM_TRANSPOSE4_PS(ewtabF,ewtabD,ewtabV,ewtabFn);
 660             felec            = _mm_add_ps(ewtabF,_mm_mul_ps(eweps,ewtabD));
 661             velec            = _mm_sub_ps(ewtabV,_mm_mul_ps(_mm_mul_ps(ewtabhalfspace,eweps),_mm_add_ps(ewtabF,felec)));
 662             velec            = _mm_mul_ps(qq20,_mm_sub_ps(rinv20,velec));
 663             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
 664
 665             /* Update potential sum for this i atom from the interaction with this j atom. */
 666             velec            = _mm_andnot_ps(dummy_mask,velec);
 667             velecsum         = _mm_add_ps(velecsum,velec);
 668
 669             fscal            = felec;
 670
 671             fscal            = _mm_andnot_ps(dummy_mask,fscal);
 672
 673             /* Calculate temporary vectorial force */
 674             tx               = _mm_mul_ps(fscal,dx20);
 675             ty               = _mm_mul_ps(fscal,dy20);
 676             tz               = _mm_mul_ps(fscal,dz20);
 677
 678             /* Update vectorial force */
 679             fix2             = _mm_add_ps(fix2,tx);
 680             fiy2             = _mm_add_ps(fiy2,ty);
 681             fiz2             = _mm_add_ps(fiz2,tz);
 682
 683             fjx0             = _mm_add_ps(fjx0,tx);
 684             fjy0             = _mm_add_ps(fjy0,ty);
 685             fjz0             = _mm_add_ps(fjz0,tz);
 686
 687             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
 688             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
 689             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
 690             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
 691
 692             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
 693
 694             /* Inner loop uses 160 flops */
 695         }
 696
 697         /* End of innermost loop */
 698
 699         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
 700                                               f+i_coord_offset,fshift+i_shift_offset);
 701
 702         ggid                        = gid[iidx];
 703         /* Update potential energies */
 704         gmx_mm_update_1pot_ps(velecsum,kernel_data->energygrp_elec+ggid);
 705         gmx_mm_update_1pot_ps(vvdwsum,kernel_data->energygrp_vdw+ggid);
 706
 707         /* Increment number of inner iterations */
 708         inneriter                  += j_index_end - j_index_start;
 709
 710         /* Outer loop uses 20 flops */
 711     }
 712
 713     /* Increment number of outer iterations */
 714     outeriter        += nri;
 715
 716     /* Update outer/inner flops */
 717
 718     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_VF,outeriter*20 + inneriter*160);
 719 }
 720 /*
 721  * Gromacs nonbonded kernel:   nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_single
 722  * Electrostatics interaction: Ewald
 723  * VdW interaction:            CubicSplineTable
 724  * Geometry:                   Water3-Particle
 725  * Calculate force/pot:        Force
 726  */
 727 void
 728 nb_kernel_ElecEw_VdwCSTab_GeomW3P1_F_sse4_1_single
 729                     (t_nblist                    * gmx_restrict       nlist,
 730                      rvec                        * gmx_restrict          xx,
 731                      rvec                        * gmx_restrict          ff,
 732                      t_forcerec                  * gmx_restrict          fr,
 733                      t_mdatoms                   * gmx_restrict     mdatoms,
 734                      nb_kernel_data_t gmx_unused * gmx_restrict kernel_data,
 735                      t_nrnb                      * gmx_restrict        nrnb)
 736 {
 737     /* Suffixes 0,1,2,3 refer to particle indices for waters in the inner or outer loop, or
 738      * just 0 for non-waters.
 739      * Suffixes A,B,C,D refer to j loop unrolling done with SSE, e.g. for the four different
 740      * jnr indices corresponding to data put in the four positions in the SIMD register.
 741      */
 742     int              i_shift_offset,i_coord_offset,outeriter,inneriter;
 743     int              j_index_start,j_index_end,jidx,nri,inr,ggid,iidx;
 744     int              jnrA,jnrB,jnrC,jnrD;
 745     int              jnrlistA,jnrlistB,jnrlistC,jnrlistD;
 746     int              j_coord_offsetA,j_coord_offsetB,j_coord_offsetC,j_coord_offsetD;
 747     int              *iinr,*jindex,*jjnr,*shiftidx,*gid;
 748     real             rcutoff_scalar;
 749     real             *shiftvec,*fshift,*x,*f;
 750     real             *fjptrA,*fjptrB,*fjptrC,*fjptrD;
 751     real             scratch[4*DIM];
 752     __m128           tx,ty,tz,fscal,rcutoff,rcutoff2,jidxall;
 753     int              vdwioffset0;
 754     __m128           ix0,iy0,iz0,fix0,fiy0,fiz0,iq0,isai0;
 755     int              vdwioffset1;
 756     __m128           ix1,iy1,iz1,fix1,fiy1,fiz1,iq1,isai1;
 757     int              vdwioffset2;
 758     __m128           ix2,iy2,iz2,fix2,fiy2,fiz2,iq2,isai2;
 759     int              vdwjidx0A,vdwjidx0B,vdwjidx0C,vdwjidx0D;
 760     __m128           jx0,jy0,jz0,fjx0,fjy0,fjz0,jq0,isaj0;
 761     __m128           dx00,dy00,dz00,rsq00,rinv00,rinvsq00,r00,qq00,c6_00,c12_00;
 762     __m128           dx10,dy10,dz10,rsq10,rinv10,rinvsq10,r10,qq10,c6_10,c12_10;
 763     __m128           dx20,dy20,dz20,rsq20,rinv20,rinvsq20,r20,qq20,c6_20,c12_20;
 764     __m128           velec,felec,velecsum,facel,crf,krf,krf2;
 765     real             *charge;
 766     int              nvdwtype;
 767     __m128           rinvsix,rvdw,vvdw,vvdw6,vvdw12,fvdw,fvdw6,fvdw12,vvdwsum,sh_vdw_invrcut6;
 768     int              *vdwtype;
 769     real             *vdwparam;
 770     __m128           one_sixth   = _mm_set1_ps(1.0/6.0);
 771     __m128           one_twelfth = _mm_set1_ps(1.0/12.0);
 772     __m128i          vfitab;
 773     __m128i          ifour       = _mm_set1_epi32(4);
 774     __m128           rt,vfeps,vftabscale,Y,F,G,H,Heps,Fp,VV,FF;
 775     real             *vftab;
 776     __m128i          ewitab;
 777     __m128           ewtabscale,eweps,sh_ewald,ewrt,ewtabhalfspace,ewtabF,ewtabFn,ewtabD,ewtabV;
 778     real             *ewtab;
 779     __m128           dummy_mask,cutoff_mask;
 780     __m128           signbit = _mm_castsi128_ps( _mm_set1_epi32(0x80000000) );
 781     __m128           one     = _mm_set1_ps(1.0);
 782     __m128           two     = _mm_set1_ps(2.0);
 783     x                = xx[0];
 784     f                = ff[0];
 785
 786     nri              = nlist->nri;
 787     iinr             = nlist->iinr;
 788     jindex           = nlist->jindex;
 789     jjnr             = nlist->jjnr;
 790     shiftidx         = nlist->shift;
 791     gid              = nlist->gid;
 792     shiftvec         = fr->shift_vec[0];
 793     fshift           = fr->fshift[0];
 794     facel            = _mm_set1_ps(fr->epsfac);
 795     charge           = mdatoms->chargeA;
 796     nvdwtype         = fr->ntype;
 797     vdwparam         = fr->nbfp;
 798     vdwtype          = mdatoms->typeA;
 799
 800     vftab            = kernel_data->table_vdw->data;
 801     vftabscale       = _mm_set1_ps(kernel_data->table_vdw->scale);
 802
 803     sh_ewald         = _mm_set1_ps(fr->ic->sh_ewald);
 804     ewtab            = fr->ic->tabq_coul_F;
 805     ewtabscale       = _mm_set1_ps(fr->ic->tabq_scale);
 806     ewtabhalfspace   = _mm_set1_ps(0.5/fr->ic->tabq_scale);
 807
 808     /* Setup water-specific parameters */
 809     inr              = nlist->iinr[0];
 810     iq0              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+0]));
 811     iq1              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+1]));
 812     iq2              = _mm_mul_ps(facel,_mm_set1_ps(charge[inr+2]));
 813     vdwioffset0      = 2*nvdwtype*vdwtype[inr+0];
 814
 815     /* Avoid stupid compiler warnings */
 816     jnrA = jnrB = jnrC = jnrD = 0;
 817     j_coord_offsetA = 0;
 818     j_coord_offsetB = 0;
 819     j_coord_offsetC = 0;
 820     j_coord_offsetD = 0;
 821
 822     outeriter        = 0;
 823     inneriter        = 0;
 824
 825     for(iidx=0;iidx<4*DIM;iidx++)
 826     {
 827         scratch[iidx] = 0.0;
 828     }
 829
 830     /* Start outer loop over neighborlists */
 831     for(iidx=0; iidx<nri; iidx++)
 832     {
 833         /* Load shift vector for this list */
 834         i_shift_offset   = DIM*shiftidx[iidx];
 835
 836         /* Load limits for loop over neighbors */
 837         j_index_start    = jindex[iidx];
 838         j_index_end      = jindex[iidx+1];
 839
 840         /* Get outer coordinate index */
 841         inr              = iinr[iidx];
 842         i_coord_offset   = DIM*inr;
 843
 844         /* Load i particle coords and add shift vector */
 845         gmx_mm_load_shift_and_3rvec_broadcast_ps(shiftvec+i_shift_offset,x+i_coord_offset,
 846                                                  &ix0,&iy0,&iz0,&ix1,&iy1,&iz1,&ix2,&iy2,&iz2);
 847
 848         fix0             = _mm_setzero_ps();
 849         fiy0             = _mm_setzero_ps();
 850         fiz0             = _mm_setzero_ps();
 851         fix1             = _mm_setzero_ps();
 852         fiy1             = _mm_setzero_ps();
 853         fiz1             = _mm_setzero_ps();
 854         fix2             = _mm_setzero_ps();
 855         fiy2             = _mm_setzero_ps();
 856         fiz2             = _mm_setzero_ps();
 857
 858         /* Start inner kernel loop */
 859         for(jidx=j_index_start; jidx<j_index_end && jjnr[jidx+3]>=0; jidx+=4)
 860         {
 861
 862             /* Get j neighbor index, and coordinate index */
 863             jnrA             = jjnr[jidx];
 864             jnrB             = jjnr[jidx+1];
 865             jnrC             = jjnr[jidx+2];
 866             jnrD             = jjnr[jidx+3];
 867             j_coord_offsetA  = DIM*jnrA;
 868             j_coord_offsetB  = DIM*jnrB;
 869             j_coord_offsetC  = DIM*jnrC;
 870             j_coord_offsetD  = DIM*jnrD;
 871
 872             /* load j atom coordinates */
 873             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
 874                                               x+j_coord_offsetC,x+j_coord_offsetD,
 875                                               &jx0,&jy0,&jz0);
 876
 877             /* Calculate displacement vector */
 878             dx00             = _mm_sub_ps(ix0,jx0);
 879             dy00             = _mm_sub_ps(iy0,jy0);
 880             dz00             = _mm_sub_ps(iz0,jz0);
 881             dx10             = _mm_sub_ps(ix1,jx0);
 882             dy10             = _mm_sub_ps(iy1,jy0);
 883             dz10             = _mm_sub_ps(iz1,jz0);
 884             dx20             = _mm_sub_ps(ix2,jx0);
 885             dy20             = _mm_sub_ps(iy2,jy0);
 886             dz20             = _mm_sub_ps(iz2,jz0);
 887
 888             /* Calculate squared distance and things based on it */
 889             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
 890             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
 891             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
 892
 893             rinv00           = gmx_mm_invsqrt_ps(rsq00);
 894             rinv10           = gmx_mm_invsqrt_ps(rsq10);
 895             rinv20           = gmx_mm_invsqrt_ps(rsq20);
 896
 897             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
 898             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
 899             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
 900
 901             /* Load parameters for j particles */
 902             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
 903                                                               charge+jnrC+0,charge+jnrD+0);
 904             vdwjidx0A        = 2*vdwtype[jnrA+0];
 905             vdwjidx0B        = 2*vdwtype[jnrB+0];
 906             vdwjidx0C        = 2*vdwtype[jnrC+0];
 907             vdwjidx0D        = 2*vdwtype[jnrD+0];
 908
 909             fjx0             = _mm_setzero_ps();
 910             fjy0             = _mm_setzero_ps();
 911             fjz0             = _mm_setzero_ps();
 912
 913             /**************************
 914              * CALCULATE INTERACTIONS *
 915              **************************/
 916
 917             r00              = _mm_mul_ps(rsq00,rinv00);
 918
 919             /* Compute parameters for interactions between i and j atoms */
 920             qq00             = _mm_mul_ps(iq0,jq0);
 921             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
 922                                          vdwparam+vdwioffset0+vdwjidx0B,
 923                                          vdwparam+vdwioffset0+vdwjidx0C,
 924                                          vdwparam+vdwioffset0+vdwjidx0D,
 925                                          &c6_00,&c12_00);
 926
 927             /* Calculate table index by multiplying r with table scale and truncate to integer */
 928             rt               = _mm_mul_ps(r00,vftabscale);
 929             vfitab           = _mm_cvttps_epi32(rt);
 930             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
 931             vfitab           = _mm_slli_epi32(vfitab,3);
 932
 933             /* EWALD ELECTROSTATICS */
 934
 935             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 936             ewrt             = _mm_mul_ps(r00,ewtabscale);
 937             ewitab           = _mm_cvttps_epi32(ewrt);
 938             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
 939             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
 940                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
 941                                          &ewtabF,&ewtabFn);
 942             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
 943             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
 944
 945             /* CUBIC SPLINE TABLE DISPERSION */
 946             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 947             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 948             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 949             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 950             _MM_TRANSPOSE4_PS(Y,F,G,H);
 951             Heps             = _mm_mul_ps(vfeps,H);
 952             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 953             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 954             fvdw6            = _mm_mul_ps(c6_00,FF);
 955
 956             /* CUBIC SPLINE TABLE REPULSION */
 957             vfitab           = _mm_add_epi32(vfitab,ifour);
 958             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
 959             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
 960             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
 961             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
 962             _MM_TRANSPOSE4_PS(Y,F,G,H);
 963             Heps             = _mm_mul_ps(vfeps,H);
 964             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
 965             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
 966             fvdw12           = _mm_mul_ps(c12_00,FF);
 967             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
 968
 969             fscal            = _mm_add_ps(felec,fvdw);
 970
 971             /* Calculate temporary vectorial force */
 972             tx               = _mm_mul_ps(fscal,dx00);
 973             ty               = _mm_mul_ps(fscal,dy00);
 974             tz               = _mm_mul_ps(fscal,dz00);
 975
 976             /* Update vectorial force */
 977             fix0             = _mm_add_ps(fix0,tx);
 978             fiy0             = _mm_add_ps(fiy0,ty);
 979             fiz0             = _mm_add_ps(fiz0,tz);
 980
 981             fjx0             = _mm_add_ps(fjx0,tx);
 982             fjy0             = _mm_add_ps(fjy0,ty);
 983             fjz0             = _mm_add_ps(fjz0,tz);
 984
 985             /**************************
 986              * CALCULATE INTERACTIONS *
 987              **************************/
 988
 989             r10              = _mm_mul_ps(rsq10,rinv10);
 990
 991             /* Compute parameters for interactions between i and j atoms */
 992             qq10             = _mm_mul_ps(iq1,jq0);
 993
 994             /* EWALD ELECTROSTATICS */
 995
 996             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
 997             ewrt             = _mm_mul_ps(r10,ewtabscale);
 998             ewitab           = _mm_cvttps_epi32(ewrt);
 999             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1000             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1001                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1002                                          &ewtabF,&ewtabFn);
1003             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1004             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1005
1006             fscal            = felec;
1007
1008             /* Calculate temporary vectorial force */
1009             tx               = _mm_mul_ps(fscal,dx10);
1010             ty               = _mm_mul_ps(fscal,dy10);
1011             tz               = _mm_mul_ps(fscal,dz10);
1012
1013             /* Update vectorial force */
1014             fix1             = _mm_add_ps(fix1,tx);
1015             fiy1             = _mm_add_ps(fiy1,ty);
1016             fiz1             = _mm_add_ps(fiz1,tz);
1017
1018             fjx0             = _mm_add_ps(fjx0,tx);
1019             fjy0             = _mm_add_ps(fjy0,ty);
1020             fjz0             = _mm_add_ps(fjz0,tz);
1021
1022             /**************************
1023              * CALCULATE INTERACTIONS *
1024              **************************/
1025
1026             r20              = _mm_mul_ps(rsq20,rinv20);
1027
1028             /* Compute parameters for interactions between i and j atoms */
1029             qq20             = _mm_mul_ps(iq2,jq0);
1030
1031             /* EWALD ELECTROSTATICS */
1032
1033             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1034             ewrt             = _mm_mul_ps(r20,ewtabscale);
1035             ewitab           = _mm_cvttps_epi32(ewrt);
1036             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1037             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1038                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1039                                          &ewtabF,&ewtabFn);
1040             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1041             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1042
1043             fscal            = felec;
1044
1045             /* Calculate temporary vectorial force */
1046             tx               = _mm_mul_ps(fscal,dx20);
1047             ty               = _mm_mul_ps(fscal,dy20);
1048             tz               = _mm_mul_ps(fscal,dz20);
1049
1050             /* Update vectorial force */
1051             fix2             = _mm_add_ps(fix2,tx);
1052             fiy2             = _mm_add_ps(fiy2,ty);
1053             fiz2             = _mm_add_ps(fiz2,tz);
1054
1055             fjx0             = _mm_add_ps(fjx0,tx);
1056             fjy0             = _mm_add_ps(fjy0,ty);
1057             fjz0             = _mm_add_ps(fjz0,tz);
1058
1059             fjptrA             = f+j_coord_offsetA;
1060             fjptrB             = f+j_coord_offsetB;
1061             fjptrC             = f+j_coord_offsetC;
1062             fjptrD             = f+j_coord_offsetD;
1063
1064             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1065
1066             /* Inner loop uses 134 flops */
1067         }
1068
1069         if(jidx<j_index_end)
1070         {
1071
1072             /* Get j neighbor index, and coordinate index */
1073             jnrlistA         = jjnr[jidx];
1074             jnrlistB         = jjnr[jidx+1];
1075             jnrlistC         = jjnr[jidx+2];
1076             jnrlistD         = jjnr[jidx+3];
1077             /* Sign of each element will be negative for non-real atoms.
1078              * This mask will be 0xFFFFFFFF for dummy entries and 0x0 for real ones,
1079              * so use it as val = _mm_andnot_ps(mask,val) to clear dummy entries.
1080              */
1081             dummy_mask = gmx_mm_castsi128_ps(_mm_cmplt_epi32(_mm_loadu_si128((const __m128i *)(jjnr+jidx)),_mm_setzero_si128()));
1082             jnrA       = (jnrlistA>=0) ? jnrlistA : 0;
1083             jnrB       = (jnrlistB>=0) ? jnrlistB : 0;
1084             jnrC       = (jnrlistC>=0) ? jnrlistC : 0;
1085             jnrD       = (jnrlistD>=0) ? jnrlistD : 0;
1086             j_coord_offsetA  = DIM*jnrA;
1087             j_coord_offsetB  = DIM*jnrB;
1088             j_coord_offsetC  = DIM*jnrC;
1089             j_coord_offsetD  = DIM*jnrD;
1090
1091             /* load j atom coordinates */
1092             gmx_mm_load_1rvec_4ptr_swizzle_ps(x+j_coord_offsetA,x+j_coord_offsetB,
1093                                               x+j_coord_offsetC,x+j_coord_offsetD,
1094                                               &jx0,&jy0,&jz0);
1095
1096             /* Calculate displacement vector */
1097             dx00             = _mm_sub_ps(ix0,jx0);
1098             dy00             = _mm_sub_ps(iy0,jy0);
1099             dz00             = _mm_sub_ps(iz0,jz0);
1100             dx10             = _mm_sub_ps(ix1,jx0);
1101             dy10             = _mm_sub_ps(iy1,jy0);
1102             dz10             = _mm_sub_ps(iz1,jz0);
1103             dx20             = _mm_sub_ps(ix2,jx0);
1104             dy20             = _mm_sub_ps(iy2,jy0);
1105             dz20             = _mm_sub_ps(iz2,jz0);
1106
1107             /* Calculate squared distance and things based on it */
1108             rsq00            = gmx_mm_calc_rsq_ps(dx00,dy00,dz00);
1109             rsq10            = gmx_mm_calc_rsq_ps(dx10,dy10,dz10);
1110             rsq20            = gmx_mm_calc_rsq_ps(dx20,dy20,dz20);
1111
1112             rinv00           = gmx_mm_invsqrt_ps(rsq00);
1113             rinv10           = gmx_mm_invsqrt_ps(rsq10);
1114             rinv20           = gmx_mm_invsqrt_ps(rsq20);
1115
1116             rinvsq00         = _mm_mul_ps(rinv00,rinv00);
1117             rinvsq10         = _mm_mul_ps(rinv10,rinv10);
1118             rinvsq20         = _mm_mul_ps(rinv20,rinv20);
1119
1120             /* Load parameters for j particles */
1121             jq0              = gmx_mm_load_4real_swizzle_ps(charge+jnrA+0,charge+jnrB+0,
1122                                                               charge+jnrC+0,charge+jnrD+0);
1123             vdwjidx0A        = 2*vdwtype[jnrA+0];
1124             vdwjidx0B        = 2*vdwtype[jnrB+0];
1125             vdwjidx0C        = 2*vdwtype[jnrC+0];
1126             vdwjidx0D        = 2*vdwtype[jnrD+0];
1127
1128             fjx0             = _mm_setzero_ps();
1129             fjy0             = _mm_setzero_ps();
1130             fjz0             = _mm_setzero_ps();
1131
1132             /**************************
1133              * CALCULATE INTERACTIONS *
1134              **************************/
1135
1136             r00              = _mm_mul_ps(rsq00,rinv00);
1137             r00              = _mm_andnot_ps(dummy_mask,r00);
1138
1139             /* Compute parameters for interactions between i and j atoms */
1140             qq00             = _mm_mul_ps(iq0,jq0);
1141             gmx_mm_load_4pair_swizzle_ps(vdwparam+vdwioffset0+vdwjidx0A,
1142                                          vdwparam+vdwioffset0+vdwjidx0B,
1143                                          vdwparam+vdwioffset0+vdwjidx0C,
1144                                          vdwparam+vdwioffset0+vdwjidx0D,
1145                                          &c6_00,&c12_00);
1146
1147             /* Calculate table index by multiplying r with table scale and truncate to integer */
1148             rt               = _mm_mul_ps(r00,vftabscale);
1149             vfitab           = _mm_cvttps_epi32(rt);
1150             vfeps            = _mm_sub_ps(rt,_mm_round_ps(rt, _MM_FROUND_FLOOR));
1151             vfitab           = _mm_slli_epi32(vfitab,3);
1152
1153             /* EWALD ELECTROSTATICS */
1154
1155             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1156             ewrt             = _mm_mul_ps(r00,ewtabscale);
1157             ewitab           = _mm_cvttps_epi32(ewrt);
1158             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1159             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1160                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1161                                          &ewtabF,&ewtabFn);
1162             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1163             felec            = _mm_mul_ps(_mm_mul_ps(qq00,rinv00),_mm_sub_ps(rinvsq00,felec));
1164
1165             /* CUBIC SPLINE TABLE DISPERSION */
1166             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1167             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1168             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1169             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1170             _MM_TRANSPOSE4_PS(Y,F,G,H);
1171             Heps             = _mm_mul_ps(vfeps,H);
1172             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1173             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1174             fvdw6            = _mm_mul_ps(c6_00,FF);
1175
1176             /* CUBIC SPLINE TABLE REPULSION */
1177             vfitab           = _mm_add_epi32(vfitab,ifour);
1178             Y                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,0) );
1179             F                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,1) );
1180             G                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,2) );
1181             H                = _mm_load_ps( vftab + gmx_mm_extract_epi32(vfitab,3) );
1182             _MM_TRANSPOSE4_PS(Y,F,G,H);
1183             Heps             = _mm_mul_ps(vfeps,H);
1184             Fp               = _mm_add_ps(F,_mm_mul_ps(vfeps,_mm_add_ps(G,Heps)));
1185             FF               = _mm_add_ps(Fp,_mm_mul_ps(vfeps,_mm_add_ps(G,_mm_add_ps(Heps,Heps))));
1186             fvdw12           = _mm_mul_ps(c12_00,FF);
1187             fvdw             = _mm_xor_ps(signbit,_mm_mul_ps(_mm_add_ps(fvdw6,fvdw12),_mm_mul_ps(vftabscale,rinv00)));
1188
1189             fscal            = _mm_add_ps(felec,fvdw);
1190
1191             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1192
1193             /* Calculate temporary vectorial force */
1194             tx               = _mm_mul_ps(fscal,dx00);
1195             ty               = _mm_mul_ps(fscal,dy00);
1196             tz               = _mm_mul_ps(fscal,dz00);
1197
1198             /* Update vectorial force */
1199             fix0             = _mm_add_ps(fix0,tx);
1200             fiy0             = _mm_add_ps(fiy0,ty);
1201             fiz0             = _mm_add_ps(fiz0,tz);
1202
1203             fjx0             = _mm_add_ps(fjx0,tx);
1204             fjy0             = _mm_add_ps(fjy0,ty);
1205             fjz0             = _mm_add_ps(fjz0,tz);
1206
1207             /**************************
1208              * CALCULATE INTERACTIONS *
1209              **************************/
1210
1211             r10              = _mm_mul_ps(rsq10,rinv10);
1212             r10              = _mm_andnot_ps(dummy_mask,r10);
1213
1214             /* Compute parameters for interactions between i and j atoms */
1215             qq10             = _mm_mul_ps(iq1,jq0);
1216
1217             /* EWALD ELECTROSTATICS */
1218
1219             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1220             ewrt             = _mm_mul_ps(r10,ewtabscale);
1221             ewitab           = _mm_cvttps_epi32(ewrt);
1222             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1223             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1224                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1225                                          &ewtabF,&ewtabFn);
1226             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1227             felec            = _mm_mul_ps(_mm_mul_ps(qq10,rinv10),_mm_sub_ps(rinvsq10,felec));
1228
1229             fscal            = felec;
1230
1231             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1232
1233             /* Calculate temporary vectorial force */
1234             tx               = _mm_mul_ps(fscal,dx10);
1235             ty               = _mm_mul_ps(fscal,dy10);
1236             tz               = _mm_mul_ps(fscal,dz10);
1237
1238             /* Update vectorial force */
1239             fix1             = _mm_add_ps(fix1,tx);
1240             fiy1             = _mm_add_ps(fiy1,ty);
1241             fiz1             = _mm_add_ps(fiz1,tz);
1242
1243             fjx0             = _mm_add_ps(fjx0,tx);
1244             fjy0             = _mm_add_ps(fjy0,ty);
1245             fjz0             = _mm_add_ps(fjz0,tz);
1246
1247             /**************************
1248              * CALCULATE INTERACTIONS *
1249              **************************/
1250
1251             r20              = _mm_mul_ps(rsq20,rinv20);
1252             r20              = _mm_andnot_ps(dummy_mask,r20);
1253
1254             /* Compute parameters for interactions between i and j atoms */
1255             qq20             = _mm_mul_ps(iq2,jq0);
1256
1257             /* EWALD ELECTROSTATICS */
1258
1259             /* Calculate Ewald table index by multiplying r with scale and truncate to integer */
1260             ewrt             = _mm_mul_ps(r20,ewtabscale);
1261             ewitab           = _mm_cvttps_epi32(ewrt);
1262             eweps            = _mm_sub_ps(ewrt,_mm_round_ps(ewrt, _MM_FROUND_FLOOR));
1263             gmx_mm_load_4pair_swizzle_ps(ewtab + gmx_mm_extract_epi32(ewitab,0),ewtab + gmx_mm_extract_epi32(ewitab,1),
1264                                          ewtab + gmx_mm_extract_epi32(ewitab,2),ewtab + gmx_mm_extract_epi32(ewitab,3),
1265                                          &ewtabF,&ewtabFn);
1266             felec            = _mm_add_ps(_mm_mul_ps( _mm_sub_ps(one,eweps),ewtabF),_mm_mul_ps(eweps,ewtabFn));
1267             felec            = _mm_mul_ps(_mm_mul_ps(qq20,rinv20),_mm_sub_ps(rinvsq20,felec));
1268
1269             fscal            = felec;
1270
1271             fscal            = _mm_andnot_ps(dummy_mask,fscal);
1272
1273             /* Calculate temporary vectorial force */
1274             tx               = _mm_mul_ps(fscal,dx20);
1275             ty               = _mm_mul_ps(fscal,dy20);
1276             tz               = _mm_mul_ps(fscal,dz20);
1277
1278             /* Update vectorial force */
1279             fix2             = _mm_add_ps(fix2,tx);
1280             fiy2             = _mm_add_ps(fiy2,ty);
1281             fiz2             = _mm_add_ps(fiz2,tz);
1282
1283             fjx0             = _mm_add_ps(fjx0,tx);
1284             fjy0             = _mm_add_ps(fjy0,ty);
1285             fjz0             = _mm_add_ps(fjz0,tz);
1286
1287             fjptrA             = (jnrlistA>=0) ? f+j_coord_offsetA : scratch;
1288             fjptrB             = (jnrlistB>=0) ? f+j_coord_offsetB : scratch;
1289             fjptrC             = (jnrlistC>=0) ? f+j_coord_offsetC : scratch;
1290             fjptrD             = (jnrlistD>=0) ? f+j_coord_offsetD : scratch;
1291
1292             gmx_mm_decrement_1rvec_4ptr_swizzle_ps(fjptrA,fjptrB,fjptrC,fjptrD,fjx0,fjy0,fjz0);
1293
1294             /* Inner loop uses 137 flops */
1295         }
1296
1297         /* End of innermost loop */
1298
1299         gmx_mm_update_iforce_3atom_swizzle_ps(fix0,fiy0,fiz0,fix1,fiy1,fiz1,fix2,fiy2,fiz2,
1300                                               f+i_coord_offset,fshift+i_shift_offset);
1301
1302         /* Increment number of inner iterations */
1303         inneriter                  += j_index_end - j_index_start;
1304
1305         /* Outer loop uses 18 flops */
1306     }
1307
1308     /* Increment number of outer iterations */
1309     outeriter        += nri;
1310
1311     /* Update outer/inner flops */
1312
1313     inc_nrnb(nrnb,eNR_NBKERNEL_ELEC_VDW_W3_F,outeriter*18 + inneriter*137);
1314 }